[lab] Victor · Data & AI Engineer — data platforms, agents, quant & product
I build systems that think, move data and become product.
Sync engines and dashboards that leadership trusts, forensics on databases that fail silently, an optimizer that knows the macro regime, and LLMs running on a laptop GPU.
01 /what I do
0
Agents & LLMs
- → Adapting an open-source multi-agent graph to my own stack
- → Quantized LLMs on an 8 GB laptop GPU and a rented 48 GB GPU
- → A working harness for coding agents on production data
0
Quant & Research
- → A Markowitz optimizer that knows the macro regime
- → Event studies turned into dampened expected-return adjustments
- → A research terminal where optimizer and agents disagree in public
0
Data Engineering
- → Sync engines with checksums, self-chaining and atomic publish
- → Forensics on a legacy MySQL whose failures were silent
- → Ingestion, attribution and QA for marketing and finance data
0
Product & Frontend
- → A hub of production dashboards that replaced a legacy BI tool
- → Tiered views where hidden numbers can't be recomputed
- → From reproduce-first migration to a clone-and-go scaffold
0
Auth & Security
- → Google sign-in with a domain allowlist and roles per dashboard
- → Postgres function exposure measured with curl, not described
- → Role guards inside server functions, not only in the UI
02 /featured projects
Python ·TypeScript ·SQL ·Postgres ·MySQL ·Supabase ·AWS Glue/Athena/S3 ·Iceberg ·Terraform ·DuckDB ·React ·TanStack ·Recharts ·Playwright ·LangGraph ·llama.cpp ·vLLM ·MCP ·Claude Code ·Docker ·Cloudflare Workers ·Next.js ·ChromaDB ·Python ·TypeScript ·SQL ·Postgres ·MySQL ·Supabase ·AWS Glue/Athena/S3 ·Iceberg ·Terraform ·DuckDB ·React ·TanStack ·Recharts ·Playwright ·LangGraph ·llama.cpp ·vLLM ·MCP ·Claude Code ·Docker ·Cloudflare Workers ·Next.js ·ChromaDB ·
03 /how I work
01 /
Measure first
Query the source before claiming anything — most “missing” data is already somewhere.
02 /
Contract / spec
Definitions, grain and acceptance numbers agreed before code.
03 /
Agent executes
Coding agents do the work with read-only access and a memory of past incidents.
04 /
Invariants
Cross-layer checks that fail loudly, each seeded from a real bug.
05 /
Human review
Access, deploys and pushes wait for an explicit OK.
→ let's talk